CN102608916A - Cell-machine based dynamic scheduling method for large part flexible job shop - Google Patents

Abstract

The invention relates to a cell-machine based dynamic scheduling method for a large part flexible job shop, which comprises the steps of: 1) establishing a gridding model of a cell machine; 2) describing the cell state; 3) defining initial and entrance boundary conditions of a whole model system as beats of particles entering the system, wherein the exit boundary condition is reception capacity of a downriver factory of a supply chain on products; 4) determining the self-organized evolution rule of the model; and 5) for the work station selecting rule R[p--c], taking completion of the jth step as the start point and completion of the (j+1)th step as the finish point, taking the period (wq+w1/pe) as a measuring standard, selecting the work station with minimum (wq+w1/pe) as the processing work station of the (j+1)th step, wherein the work sequencing rule R[c--p] is determined according to a ''first come first serviced'' rule with the combination of workpiece processing priority. According to the cell-machine based dynamic scheduling method for the large part flexible job shop, the operating efficiency is improved and the stability of the device is enhanced.

Description

The dynamic dispatching method of the flexible job shop of a kind of heavy parts based on cellular machine

Technical field

The present invention relates to the dispatching method of job shop, the dynamic dispatching method of the flexible job shop of especially a kind of heavy parts.

Background technology

Along with the fast development of science and technology and being growing more intense of market competition; The production model of most of enterprise is changed to many kinds, short run or single-piece production by traditional single, production in enormous quantities; The production run structure type is further complicated, and traditional production scheduling method can't adapt to the order demand of height change.Be in the price earning ratio and the Unpredictability of colony of the manufacturer order between supplier and the storage merchant on the supply chain, make its production scheduling problems become especially complicated.

The production scheduling of heavy parts workshop on the supply chain; Belong to one type of scheduling that flexible job shop scheduling (Flexible Job-Shop Scheduling) and dynamic dispatching (Dynamic Scheduling) combine, and characteristics such as the workpiece volume is big, machining period is long are arranged.The subject matter of the conventional scheduling of this type of enterprise at present is:

(1) shop equipment load is uneven, and equipment component can't running at full capacity, even with than low load operation, the equipment component overload operation;

(2) ability that reply equipment failure, urgent slotting list, workpiece are reprocessed, the portioned product order can't deliver goods on schedule;

(3) a little less than the production capacity computational estimation competence of equipment failure rate and individual device.

At present, considerably less to the research of big machinery part workshop scheduling problem fully both at home and abroad, also few about the research of flexible job shop dynamic dispatching, the same with general production scheduling problems.

Summary of the invention

The deficiency of, less stable lower for the equipment operating efficiency that overcomes existing existing job shop dispatching method, the present invention provides a kind of dynamic dispatching method based on the flexible job shop of heavy parts of cellular machine that improves equipment operating efficiency, enhanced stability.

The technical solution adopted for the present invention to solve the technical problems is:

The dynamic dispatching method of the flexible job shop of a kind of heavy parts based on cellular machine comprises the steps:

1) set up the grid model of cellular machine: whole workshop is set at one comprises improved two-dimensional grid system, each grid is represented a processing stations or buffer memory as a cellular, and a certain moment of each station can only be processed a part; The workpiece of buffer area is waited in line; Set total n the station group of the flexible job shop of said heavy parts, every group of station number m (n) that comprises is only relevant with station group alias n, does not generally wait; The station number that each station group comprises with the buffer memory number of its corresponding cache group mutually;, improved expression arrives the workpiece of system at random

2) cellular state is described design: can represent (1) with following mathematical form according to any buffer memory cellular t+1 of cellular state function

state constantly; Station cellular t+1 STA representation constantly is (2) equally, arbitrarily; Buffer memory cellular t status attribute constantly is expressed as (3); Station cellular t status attribute constantly is expressed as (4):

3) starting condition and boundary condition are set: workpiece particle t status attribute constantly; Expression formula is (5) as follows; The initial sum inlet boundary conditional definition of The model system becomes each particle to get into the beat of system, and export boundary condition is the ability to accept of supply chain downstream factory to product;

4) confirm model self-organization evolution rule: the optimization aim of the flexible job shop dynamic scheduling problem of heavy parts is summarised as 3 points: (1) each procedure completion date early; (2) each station rate of load condensate is high, and is idle few; (3) all station balancing the loads in the same station group reduce three according to the characteristics of flexible job shop dynamic dispatching with the self-organization evolution rule of this model, shown in the table 1,

Table 1

4) model evolution process specifically comprises:

Station is selected: station selective rule R _{P → c}: a certain workpiece arrives, and at first judges busy, the spare time or the malfunction of station in its next target station group; According to the quantity that is in the station of same state, preferentially selecting the working (machining) efficiency height also is the big station of pe numerical value, or can accomplish the station of this operation the earliest, promptly compares (w then _q+ wl ÷ pe) value, the station of the correspondence that selective value is minimum; Judge again whether workpiece can be held by its target station cellular and target cache cellular on time, space, also promptly compare s _ClWith the size of wl, and w _ClAnd s _nSize, if s _Cl>=wl and w _Cl>=s _n, workpiece gets into station, otherwise workpiece temporarily gets into the reserve buffer area, treats that the emulation clock advances the step to rejudge when dispatching one the time again;

Workpiece sequencing: workpiece sequencing rule R _{C → p}The processing priority of main combination FCFS (First Come First Served) principle and workpiece is confirmed the processing sequence of workpiece, at first, and according to due in t _aOrdering, t _aMore little workpiece comes front more; The workpiece size of priority dp relatively in twos forward from the team end then; If the dp value of back workpiece is greater than the dp value of front workpiece; Then queue position and the formation sequence number qn with two workpiece exchanges, and is not less than till the dp value of back workpiece the output queue order up to the dp of front workpiece value; Upgrade the state attribute value of buffer memory cellular and each particle

Task triggers: task triggering rule R _TaActivation need satisfy two conditions: (1) target station cellular is idle, s _s=0; (2) workpiece to be processed is positioned at formation foremost, qn=1, and when condition satisfies, workpiece target approach station, the target station is occupied, and puts it and has much to do s _s=1, buffer memory cellular queue length lq=lq-1, the formation sequence number qn of all particles of waiting in the formation subtracts 1, advances one, produces the processing start time, upgrades t simultaneously _a, arrange the process finishing incident after the completion of processing, idle station, workpiece leaves, and returns executive program at last,

Evolution rule is: station selective rule R _{P → c}Be that to machine with the j procedure be starting point, the j+1 procedure machines the (w during this period of time into terminal point _q+ wl ÷ pe) is criterion, selects (w _q+ wl ÷ pe) minimum station is as the processing stations of j+1 procedure, workpiece sequencing rule R _{C → p}Then combine the decision of workpiece processing priority according to the First Come First Served rule.

Further, in the said step 4), adopt genetic algorithm that the evolution rule of cellular machine is optimized, process is following:

If station group is a unit with its corresponding cache group, each the time step in, the scheduling mechanism of each unit is identical, is converted into static one by one flexible scheduling problem, optimal objective has 3: (1) each procedure completion date early; (2) each station rate of load condensate is high, and is idle few; (3) all station balancing the loads in the same station group, below, utilize genetic algorithm to station selective rule R _{P → c}With workpiece sequencing rule R _{C → p}Encode;

The model description of each scheduling unit of cellular machine in each time step is: n workpiece particle will be processed in the station group that comprises the individual similar station of m; M the corresponding m of a station difference buffer memory; The working (machining) efficiency of each station is different, and each workpiece only needs to accomplish one procedure in this station group, and to require alternative station be that in the station group certain is several or whole to the per pass operation according to actual; The per pass operation is processed the required time and is differed on different stations, establish:

Workpiece particle collection P={p ₁, p ₂, p ₃..., p _n;

Station collection s={s ₁, s ₂, s ₃..., s _m;

Buffer memory collection B={b ₁, b ₂, b ₃..., b _m;

Operation O _IjEach workpiece has only unique one procedure in the j procedure of representing i workpiece, this model investigation scope, also is that the corresponding j of each workpiece has only one;

Matrix T process time of x station group, wherein T _IjxyRepresent that the j procedure of i workpiece processes the needed time on y station of x station group;

The ME temporal information, every rate of breakdown is λ _y

Sub-goal weight w ₁, w ₂, w ₃,

All process steps is the shortest total process time; Promptly

is in order to cooperate all the other two sub-goals; Make it relatively uniform, it is handled as follows, make sub-goal 1 be converted into maximal value; And be unlikely to too to approach 0

${\mathrm{<figure-callout\; id="1"\; label="F"\; filenames="HDA0000135858980000012.png,HDA0000135858980000013.png"\; state="\{\{state\}\}">F</figure-callout>}}_1=\frac{\underset{i=1}{\overset{n}{\mathrm{\&Sigma;}}}\min T_{\mathrm{iy}}}{\underset{i=1}{\overset{n}{\mathrm{\&Sigma;}}}\left(T_i{A}_i\right)}y\&Element;\{\mathrm{1,2},...,m\}---\left(6\right)$

In the formula (6), what divide subrepresentation is that all workpiece all select process time the shortest station to add total process time in man-hour, is the minimum value of ideal state; Denominator is represented actual total process time of each feasible solution, 0＜F ₁≤1, T _iBe the row matrix of a series of m of comprising element, the corresponding T of each element _Ijxy, can be abbreviated as T for each cell scheduling _Iy

A _iThen be the column matrix of a series of m of comprising element, each element a _Iy∈ 0,1}, wherein y is corresponding be exactly station sequence number y ∈ 1,2 ... m},

Each station rate of load condensate is high, is converted into all station overall utilization F ₂High:

${\mathrm{<figure-callout\; id="2"\; label="F"\; filenames="HDA0000135858980000012.png,HDA0000135858980000021.png"\; state="\{\{state\}\}">F</figure-callout>}}_2=\frac{\underset{i=1}{\overset{n}{\mathrm{\&Sigma;}}}\left(T_i{A}_i\right)}{\underset{y=1}{\overset{m}{\mathrm{\&Sigma;}}}\left(T(1-{\mathrm{\&lambda;}}_y)\right)}---\left(7\right)$

In the formula (7), T is the real time of each simulation step length representative, sets before emulation begins, and does not change in the simulation process; λ _yBe every rate of breakdown in the corresponding station group,

All station balancing the loads in the same station group, balanced ratio F ₃Maximum, according to

${\mathrm{<figure-callout\; id="3"\; label="F"\; filenames="HDA0000135858980000012.png,HDA0000135858980000021.png"\; state="\{\{state\}\}">F</figure-callout>}}_3=\frac{\underset{i=1}{\overset{n}{\mathrm{\&Sigma;}}}\left(T_i{A}_i\right)}{C_{\max }\&times;m}---\left(8\right)$

In the formula (8), C _MaxFor loading maximum spent process time of station, m is the station sum of station group for this reason,

Through weight add legal each discrete unit that obtains cellular machine each the time scheduling general objective function in the step:

F=max (w ₁F ₁+ w ₂F ₂+ w ₃F ₃), promptly

$F=\max ({\mathrm{<figure-callout\; id="1"\; label="w"\; filenames="HDA0000135858980000012.png,HDA0000135858980000013.png"\; state="\{\{state\}\}">w</figure-callout>}}_1\&times;\frac{\underset{i=1}{\overset{n}{\mathrm{\&Sigma;}}}\min T_{\mathrm{iy}}}{\underset{i=1}{\overset{n}{\mathrm{\&Sigma;}}}\left(T_i{A}_i\right)}+{\mathrm{<figure-callout\; id="2"\; label="w"\; filenames="HDA0000135858980000012.png,HDA0000135858980000021.png"\; state="\{\{state\}\}">w</figure-callout>}}_2\&times;\frac{\underset{i=1}{\overset{n}{\mathrm{\&Sigma;}}}\left(T_i{A}_i\right)}{\underset{y=1}{\overset{m}{\mathrm{\&Sigma;}}}\left(T(1-{\mathrm{\&lambda;}}_y)\right)}+{\mathrm{<figure-callout\; id="3"\; label="w"\; filenames="HDA0000135858980000012.png,HDA0000135858980000021.png"\; state="\{\{state\}\}">w</figure-callout>}}_3\&times;\frac{\underset{i=1}{\overset{n}{\mathrm{\&Sigma;}}}\left(T_i{A}_i\right)}{C_{\max }\&times;m})---\left(9\right)$

Processing constraint condition is following:

1) resource constraint,

A) same equipment synchronization can only be processed a workpiece;

B) the same one procedure of same workpiece can only be by an apparatus processing;

C) the per pass operation of each workpiece is just Once you begin processed and can not be stopped,

2) process constraint,

A) not constraint successively between the operation of different workpieces;

B) constraint is successively arranged between the operation of same workpiece,

3) weight constraints,

A) weight of each sub-goal is between 0 and 1: 0≤w _i≤1i ∈ 1,2,3};

B) the weight sum of all sub-goals is

Obtain initial population: with min (w _q+ wl ÷ pe) the station selection scheme that obtains for the station choice criteria; And, obtain the initial solution of genetic algorithm according to station selective staining body and operation ordering chromosome in corresponding respectively coding of the regular workpiece sequencing scheme that combines workpiece processing priority to obtain of FCFS;

Objective function is made as fitness function, has

$\mathrm{fit}\left(F\right)=F=0.4\&times;\frac{\underset{i=1}{\overset{n}{\mathrm{\&Sigma;}}}\min T_{\mathrm{iy}}}{\underset{i=1}{\overset{n}{\mathrm{\&Sigma;}}}\left(T_i{A}_i\right)}+0.3\&times;\frac{\underset{i=1}{\overset{n}{\mathrm{\&Sigma;}}}\left(T_i{A}_i\right)}{\underset{y=1}{\overset{m}{\mathrm{\&Sigma;}}}\left(T(1-{\mathrm{\&lambda;}}_y)\right)}+0.3\&times;\frac{\underset{i=1}{\overset{n}{\mathrm{\&Sigma;}}}\left(T_i{A}_i\right)}{C_{\max }\&times;m}---\left(10\right)$

Adopt pro rata fitness distribution method, proportion of utilization is in the possibility of its filial generation reservation of probability decision of each ideal adaptation degree, if certain individuals i, its fitness is f _i, then its probability tables that is selected is shown,

$P_i=\frac{f_i}{\underset{i=1}{\overset{M}{\mathrm{\&Sigma;}}}{f}_i}---\left(11\right)$

The roulette back-and-forth method is adopted in system of selection, the ideal adaptation degree is converted in proportion chooses probability, has several individuals just wheel disc to be divided into several sectors;

Intersect: said chromosome is divided into two parts, and first's chromosome station selects part directly to adopt two point of contacts bracketing method, and counterpart intersects in twos; The chromosome of second portion---in the workpiece sequencing chromosome because the per pass operation can only occur once, if adopt two chromosomes to intersect in twos, the new chromosome of generation be not identical with parent chromosome be exactly illegal;

Variation: the chromosomal variation of first, be employed in the gene string of station selective staining body and select a position at random, concentrate at the station of this operation and select an integer unequal at random with it; Replace current gene; Separating of obtaining like this can guarantee it is feasible solution, and the method for gene location exchange is adopted in the chromosomal variation of second portion, promptly from operation ordering chromosome, selects the gene of two positions at random; Its position is exchanged; Can guarantee so identical numeral can not occur in same the operation ordering chromosome, also promptly separate feasible

The simple evolutionary process of a generation comprises obtains initial population, fitness function, selection, intersection, mutation operation; Simple evolutionary process according to a generation iterates then; Up to satisfying end condition, the chromosome that obtain this moment can be considered the optimal rules of the cellular automaton evolution that after the genetic algorithm optimizing, obtains.

Technical conceive of the present invention is: analyzes the characteristics of its job shop scheduling of big machinery manufacturing works on the supply chain, summarizes the job-shop scheduling problem (flexible job shop dynamic dispatching) of a quasi-representative, and existing problem.

According to the idea about modeling of cellular automaton, the flexible job shop dynamic scheduling problem of this class factory is carried out the abstract of cellular machine realistic model, set up the cellular machine realistic model.Mainly comprise: model space grid dividing, network node state description, initial boundary condition enactment and four parts of self-organization evolution rule design.

Wherein the self-organization evolution rule comprises initial rules foundation and rule optimization two parts, and the former adopts the evolution rule of First Come First Served combination workpiece priority in the waiting line theory, and the latter adopts genetic algorithm that initial rules is optimized, and seeks satisfactory solution.The genetic algorithm optimization rule comprises the foundation of objective function again, coding and decoding design, fitness function confirm the parts such as design of genetic operators such as selection, intersection, variation.

Beneficial effect of the present invention mainly shows: improve equipment operating efficiency, enhanced stability.

Description of drawings

Fig. 1 is a heavy parts workshop cellular machine model net trrellis diagram.

Fig. 2 is model cellular neighborhood figure, wherein, (a) is black cellular C ₃₃T state constantly (b) is black cellular C ₂₃T state constantly.

Fig. 3 is a workpiece particle classifying synoptic diagram.

Fig. 4 is the synoptic diagram of instance 1 cellular machine model.

Fig. 5 is instance 1 initial solution acquisition process figure.

Fig. 6 is the synoptic diagram that instance 1 initial solution set pair is answered wheel disc.

Fig. 7 is first's station selective staining body intersection synoptic diagram.

Fig. 8 is a second portion operation ordering chromosome POX intersection synoptic diagram.

Fig. 9 is No. 1 chromosomal variation synoptic diagram in the instance 1.

Figure 10 is the first generation evolutionary process figure of genetic algorithm initial population.

Embodiment

Below in conjunction with accompanying drawing the present invention is further described.

With reference to Fig. 1～Figure 10, the dynamic dispatching method of the flexible job shop of a kind of heavy parts based on cellular machine comprises following process:

Set up the grid model of cellular machine: whole workshop is set at one comprises improved two-dimensional grid system, each grid is represented a processing stations or buffer memory as a cellular.The a certain moment of each station can only be processed a part, and the workpiece of buffer area is waited in line.Shown in Fig. 3-1, total n station group, every group of station number m (n) that comprises is only relevant with station group alias n, does not generally wait.The station number that each station group comprises and equate with the buffer memory number of its corresponding cache group.Improved expression arrives the workpiece of system at random.

Cellular state is described design: neighborhood is confirmed: as Fig. 2 (a) (b) shown in, black cellular C ₃₃, C ₂₃T state constantly only with figure in the shade cellular relevant.

Can represent with following mathematical form according to any buffer memory cellular t+1 of cellular state function

state constantly

$S_{C_{(2n+1)j}}^{t+1}=f[S_{C_{(2n+1)j}}^t,(S_{C_{\left(2n\right)m}}^t,S_{C_{(2n+1)m}}^t,S_{C_{(2n+2)j}}^t)]---\left(1\right)$

In the formula (1):

F---local state transformation rule is the job scheduling rule;

---buffer memory cellular C _{(2n+1) j}All station cellular t of upstream station group state set constantly;

---buffer memory cellular C _{(2n+1) j}Place other cellulars of buffer memory group C _{(2n+1) m}T state set constantly;

---buffer memory cellular C _{(2n+1) j}The station cellular C corresponding in the downstream process group with it _{(2n+2) j}T state constantly.

Station cellular t+1 STA representation constantly does equally, arbitrarily

$S_{C_{\left(2n\right)j}}^{t+1}=f[S_{C_{\left(2n\right)j}}^t,(S_{C_{(2n-1)j}}^t,{\mathrm{<figure-callout\; id="2n"\; label="S\; C"\; filenames="HDA0000135858980000012.png"\; state="\{\{state\}\}">S</figure-callout>}}_{C_{}}^{{}_{2n+1}})]---\left(2\right)$

In the formula (3-2):

---station cellular C _{(2n) j}In the upstream station group with its corresponding cache cellular C _{(2n-1) j}T state constantly;

---station cellular C _{(2n) j}All buffer memory cellular t of downstream buffer memory group state set constantly.

The design of cellular state attribute: buffer memory cellular t status attribute constantly is expressed as

$S_{\mathrm{cb}}^t(w_{\mathrm{ct}},w_{\mathrm{co}},w_{\mathrm{cl}},\mathrm{lq},w_q)---\left(3\right)$

In the formula (3):

---Status of Cell Buffer, buffer memory cellular C _{(2n+1) j}At t state constantly;

w _Ct---wIP Capacity in Total, the buffer memory cellular in the product space total volume, static attribute, the unit of unit;

w _Co---WIP Capacity Occupied, occupied capacity in the buffer memory cellular, dynamic attribute, 0≤w _Co≤w _Ct

w _Cl---WIP Capacity Left, the unappropriated capacity of this buffer memory cellular, dynamic attribute, w _Cl=w _Ct-w _Co

Lq---Length of the Queue, queue length, dynamic attribute, the medium workpiece number to be processed of expression buffer memory cellular;

w _q---Waiting Time of the Queue, the stand-by period, dynamic attribute equals the time that last processing work need be waited in the workpiece formation of buffer memory cellular, the day of unit,

(wl is the corresponding required working ability of operation of part).

Pe---Processing Efficiency, working (machining) efficiency, static attribute is selected to represent operation, gets can accomplish several roads in one day and represent operation as working (machining) efficiency.The pe majority of each station does not wait;

Station cellular t status attribute constantly is expressed as:

$S_{\mathrm{cs}}^t(\mathrm{st},\mathrm{pe},T,s_s,s_{\mathrm{ct}},s_{\mathrm{co}},s_{\mathrm{cl}})---\left(4\right)$

In the formula (4):

---Status ofCell Station, station cellular C _{(2n) j}At t state constantly;

St---Station Type, affiliated station group, static attribute, st ∈ { C _2n, n is a positive integer;

The time that T---station can be used to process in dispatching cycle altogether, static attribute, the sky/d of unit;

s _s---Station Status, station busy-idle condition, dynamic attribute, s _s∈ 0,1, and 2}, 0 free time, 1 is busy, 2 faults;

s _Ct---Station Capacity in Total, the total working ability of station in dispatching cycle, static attribute, s _Ct=pe * T, the unit of unit;

s _Co---Station Capacity Occupied, the occupied working ability of station also is the load of station; Dynamic attribute, the workpiece that equals to process He processing, workpiece to be processed in the formation of corresponding buffering cellular; The summation of the working ability value that this three part needs, 0≤s _Co≤s _Ct

s _Cl---Station Capacity Left, the remaining working ability of station cellular, dynamic attribute, s _Cl=s _Ct-s _Co

Starting condition and boundary condition are set: the step be not a definite value during emulation in this model, after once again the starting point of scheduling decide according to a preceding scheduling result.For describing each time step of dynamic dispatching primary crowd of scheduling more better, introduce the notion of particle window.The step pushes away further scheduling more all only to the workpiece particle in the particle window during each emulation, is called the window particle ^[21]It is 4 types that all particles of this model are divided into according to processing stage: process, process, undressed and to be processed.The window particle comprises back 3 types.Any part of doing over again, slotting single-piece, newly arrived workpiece will go on foot the initial workpiece population of dispatching again under all will becoming jointly with the particle that the last step in a period of time still rests on the particle window for the moment.The particle of processing leaves the station cellular in that original station is continued processing until completion, is again schedule constraints.

Design workpiece particle t status attribute constantly below, expression formula is following:

$S_p^t(p_t,p_f,\mathrm{np},\mathrm{fst},\mathrm{wl},s_n,{\mathrm{dp},\mathrm{qn},t}_a)---\left(5\right)$

In the formula:

---Status of Particle, particle p _kAt t state constantly, k ∈ N ⁺, the workpiece numbering;

p _t---Procedures in Total, the operation sum of the required processing of particle, static attribute;

p _f---the process number that Procedures Finished, particle have accomplished, dynamic attribute;

Np---Next Procedure, the next process numbering of particle, dynamic attribute;

Fst---Following Station Type, particle accomplish after the current operation next station group that will get into, dynamic attribute;

Wl---Workload, the working ability of the needed station of this particle next process, dynamic attribute, unit are unit;

s _n---Space Needed, the required occupation space of this particle, static attribute, the w of corresponding buffer area _Cl

Dp---Delivery Priority, workpiece processing priority, static attribute, by the decision at delivery date of workpiece, delivery date, tighter priority was high more, and corresponding dp numerical value is big more, and dp gets positive integer;

Qn---Sequence Number of the Queue, the sequence number in the queue order, dynamic attribute, qn gets positive integer;

t _a---Arrive Time, arrive time of cellular, be divided into the time (also promptly begin process time) of the time that arrives the buffer memory cellular (also promptly begin line up time) and arrival station cellular, dynamic attribute.

The initial sum inlet boundary conditional definition of The model system becomes each particle to get into the beat of system.The dynamic of taking into account system, beat have certain randomness, but not immobilize.Export boundary condition is the ability to accept of supply chain downstream factory to product, is the root of order frequent variations in the native system, and the product ability to accept of downstream factory is the combination of dynamic and stable state.Part product demand is stable, changes little; Another part product changes along with the variation of its customer demand, influences model evolution.

The design of model self-organization evolution rule: the optimization aim of the flexible job shop dynamic scheduling problem of heavy parts can be summarized as 3 points: (1) each procedure completion date early; (2) each station rate of load condensate is high, and is idle few; (3) all station balancing the loads in the same station group.According to the characteristics of flexible job shop dynamic dispatching the self-organization evolution rule of this model is reduced three, shown in the table 1.

Table 1

Model evolution: the concrete evolutionary process of model is explained through following instance 1.Suppose that certain heavy parts workshop is made up of n station group and n buffer memory group, this example is only studied one of them unit, and is as shown in Figure 2, only studies the state variation of dash area cellular.t ₀Constantly, the state attribute value of each buffer memory cellular, station cellular is like table 2, shown in 3, during the emulation in the part flow each state attribute value of particle see table 4.

Table 2

Table 3

Table 4

Station chooser program: station selective rule R _{P → c}: a certain workpiece arrives, and at first judges busy, the spare time or the malfunction of station in its next target station group; According to the quantity that is in the station of same state, preferentially selecting the working (machining) efficiency height also is the big station of pe numerical value, or can accomplish the station of this operation the earliest, promptly compares (w then _q+ wl ÷ pe) value, the station of the correspondence that selective value is minimum; Judge again whether workpiece can be held by its target station cellular and target cache cellular on time, space, also promptly compare s _ClWith the size of wl, and w _ClAnd s _nSize, if s _Cl>=wl and w _Cl>=s _n, workpiece gets into station, otherwise workpiece temporarily gets into the reserve buffer area, treats that the emulation clock advances the step to rejudge when dispatching one the time again.

P in the instance 1 ₁Arrive at first, be at subsequent processing group C ₂The middle processing of accomplishing the 1st procedure, optional station cellular set is { C ₂₁, C ₂₂, C ₂₃, C ₂₄, C ₂₅, the s of corresponding station set _sAttribute be 0,0,0,0,0}.According to regular R _{P → c}, preferentially select the big station of pe, the pe community set is that { 0.5,1,1,1.5,2} is 2 to the maximum, selects station C ₂₅Because C ₂₅Corresponding cache station C ₁₅Queue length be 0, work as p ₁Get into C ₁₅Moment, the value of qn becomes 1, activate a task triggering rule R _Ta, so p ₁State by t ₀Constantly

Be updated to t ₁Constantly Can be considered direct entering station cellular C ₂₅Accordingly, C ₂₅State by

Be updated to

Other cellulars and particle state are constant.

Workpiece sequencing subroutine: workpiece sequencing rule R _{C → p}The processing priority of main combination FCFS (First Come First Served) principle and workpiece is confirmed the processing sequence of workpiece.At first, according to due in t _aOrdering, t _aMore little workpiece comes front more; The workpiece size of priority dp relatively in twos forward from the team end then if the dp value of back workpiece is greater than the dp value of front workpiece, then with the queue position and the formation sequence number qn exchange of two workpiece, is not less than up to the dp of front workpiece value till the dp value of workpiece at the back.The output queue order, the state attribute value of renewal buffer memory cellular and each particle.

Task triggers subroutine: task triggering rule R _TaActivation need satisfy two conditions: (1) target station cellular is idle, s _s=0; (2) workpiece to be processed is positioned at formation foremost, qn=1.When condition satisfies, workpiece target approach station, the target station is occupied, and puts it and has much to do s _s=1, buffer memory cellular queue length lq=lq-1, the formation sequence number qn of all particles of waiting in the formation subtracts 1, advances one.Produce the processing start time, upgrade t simultaneously _a, arrange the process finishing incident after the completion of processing, idle station, workpiece leaves, and returns executive program at last.

Genetic algorithm optimization evolution rule: station selective rule R _{P → c}Be that to machine with the j procedure be starting point, the j+1 procedure machines the (w during this period of time into terminal point _q+ wl ÷ pe) is criterion, selects (w _q+ wl ÷ pe) station of minimum is as the processing stations of j+1 procedure.Workpiece sequencing rule R _{C → p}Then combine the decision of workpiece processing priority according to the First Come First Served rule.As the confirming of emphasis processing stations and confirming of processing sequence of dynamic flexible scheduling, above-mentioned rule obviously is not optimum evolution rule, and the scheme that is obtained is not an optimal case yet.Therefore, the content of this joint is exactly to adopt genetic algorithm that the evolution rule of cellular machine is optimized.

If station group is a unit with its corresponding cache group, each the time step in, the scheduling mechanism of each unit is identical, is converted into static one by one flexible scheduling problem.The optimal objective of this paper has 3: (1) each procedure completion date early; (2) each station rate of load condensate is high, and is idle few; (3) all station balancing the loads in the same station group.Below, utilize genetic algorithm to station selective rule R _{P → c}With workpiece sequencing rule R _{C → p}Encode, do optimization process, seek optimum evolution rule.

Cellular machine static scheduling unit is described: each scheduling unit of cellular machine each the time model in the step can be described as: n workpiece particle will be processed in the station group that comprises m similar station, m the corresponding m of a station difference buffer memory.The working (machining) efficiency of each station is different; Each workpiece only needs to accomplish one procedure in this station group; To require alternative station be that in the station group certain is several or whole to the per pass operation according to actual, and the per pass operation is processed the required time and differed on different stations.If:

1) workpiece particle collection P={p ₁, p ₂, p ₃..., p _n;

2) station collection s={s ₁, s ₂, s ₃..., s _m;

3) buffer memory collection B={b ₁, b ₂, b ₃..., b _m;

4) operation O _IjEach workpiece has only unique one procedure in the j procedure of representing i workpiece, this model investigation scope, also is that the corresponding j of each workpiece has only one;

5) matrix T process time of x station group, wherein T _IjxyRepresent that the j procedure of i workpiece processes the needed time on y station of x station group;

6) ME temporal information, every rate of breakdown is λ _y

7) sub-goal weight w ₁, w ₂, w ₃

The objective function of aforementioned three targets correspondence is following.

All process steps is the shortest total process time; Promptly is in order to cooperate all the other two sub-goals; Make it relatively uniform; It is handled as follows, make sub-goal 1 be converted into maximal value, and be unlikely to too to approach 0.

${\mathrm{<figure-callout\; id="1"\; label="F"\; filenames="HDA0000135858980000012.png,HDA0000135858980000013.png"\; state="\{\{state\}\}">F</figure-callout>}}_1=\frac{\underset{i=1}{\overset{n}{\mathrm{\&Sigma;}}}\min T_{\mathrm{iy}}}{\underset{i=1}{\overset{n}{\mathrm{\&Sigma;}}}\left(T_i{A}_i\right)}y\&Element;\{\mathrm{1,2},...,m\}---\left(6\right)$

In the formula (6), what divide subrepresentation is that all workpiece all select process time the shortest station to add total process time in man-hour, is the minimum value of ideal state; Denominator is represented actual total process time of each feasible solution, 0＜F ₁≤1.T _iBe the row matrix of a series of m of comprising element, the corresponding T of each element _Ijxy, can be abbreviated as T for each cell scheduling _Iy

A _iThen be the column matrix of a series of m of comprising element, each element a _Iy∈ 0,1}, wherein y is corresponding be exactly station sequence number y ∈ 1,2 ... m}.

Each station rate of load condensate is high, is converted into all station overall utilization F ₂High.

${\mathrm{<figure-callout\; id="2"\; label="F"\; filenames="HDA0000135858980000012.png,HDA0000135858980000021.png"\; state="\{\{state\}\}">F</figure-callout>}}_2=\frac{\underset{i=1}{\overset{n}{\mathrm{\&Sigma;}}}\left(T_i{A}_i\right)}{\underset{y=1}{\overset{m}{\mathrm{\&Sigma;}}}\left(T(1-{\mathrm{\&lambda;}}_y)\right)}---\left(7\right)$

In the formula (7), T is the real time of each simulation step length representative, sets before emulation begins, and does not change in the simulation process; λ _yIt is every rate of breakdown in the corresponding station group.

All station balancing the loads in the same station group, balanced ratio F ₃Maximum, according to

${\mathrm{<figure-callout\; id="3"\; label="F"\; filenames="HDA0000135858980000012.png,HDA0000135858980000021.png"\; state="\{\{state\}\}">F</figure-callout>}}_3=\frac{\underset{i=1}{\overset{n}{\mathrm{\&Sigma;}}}\left(T_i{A}_i\right)}{C_{\max }\&times;m}---\left(8\right)$

In the formula (8), C _MaxFor loading maximum spent process time of station, m is the station sum of station group for this reason.

Through weight add legal each discrete unit that obtains cellular machine each the time scheduling general objective function in the step:

F=max (w ₁F ₁+ w ₂F ₂+ w ₃F ₃), promptly

$F=\max ({\mathrm{<figure-callout\; id="1"\; label="w"\; filenames="HDA0000135858980000012.png,HDA0000135858980000013.png"\; state="\{\{state\}\}">w</figure-callout>}}_1\&times;\frac{\underset{i=1}{\overset{n}{\mathrm{\&Sigma;}}}\min T_{\mathrm{iy}}}{\underset{i=1}{\overset{n}{\mathrm{\&Sigma;}}}\left(T_i{A}_i\right)}+{\mathrm{<figure-callout\; id="2"\; label="w"\; filenames="HDA0000135858980000012.png,HDA0000135858980000021.png"\; state="\{\{state\}\}">w</figure-callout>}}_2\&times;\frac{\underset{i=1}{\overset{n}{\mathrm{\&Sigma;}}}\left(T_i{A}_i\right)}{\underset{y=1}{\overset{m}{\mathrm{\&Sigma;}}}\left(T(1-{\mathrm{\&lambda;}}_y)\right)}+{\mathrm{<figure-callout\; id="3"\; label="w"\; filenames="HDA0000135858980000012.png,HDA0000135858980000021.png"\; state="\{\{state\}\}">w</figure-callout>}}_3\&times;\frac{\underset{i=1}{\overset{n}{\mathrm{\&Sigma;}}}\left(T_i{A}_i\right)}{C_{\max }\&times;m})---\left(9\right)$

Processing constraint condition is following:

1) resource constraint.

A) same equipment synchronization can only be processed a workpiece;

B) the same one procedure of same workpiece can only be by an apparatus processing;

C) the per pass operation of each workpiece is just Once you begin processed and can not be stopped.

2) process constraint.

A) not constraint successively between the operation of different workpieces;

B) constraint is successively arranged between the operation of same workpiece.

3) weight constraints.

A) weight of each sub-goal is between 0 and 1: 0≤w _i≤1i ∈ 1,2,3};

B) the weight sum of all sub-goals is

Extension with instance 1 specifies.According to the instance 1 of front, there are 8 workpiece p in the step during t of known cellular machine model emulation _n, (n=1,2 ..., 8) get into No. 2 station groups, this station group comprises 5 common apparatus s of the same type _m, (m=1,2 ..., 5), performance separately, efficient difference.Each workpiece is processed the required time shown in table 3-5 on individual device, wherein, and "---" expression workpiece p _iCan not be at station s _jLast processing, reason comprise that workpiece processing requires and equipment performance does not match, or should the time go on foot in, equipment just is in malfunction etc.

Compare instance 1, table 5 gives the corresponding operation numbering of each workpiece,, sees from the overall situation that each workpiece all comprises multiple working procedure though each workpiece has only one procedure in the cell scheduling, provides the explanation that the operation numbering is convenient to final overall scheduling scheme.

Table 5

The cellular machine model of instance 1 correspondence is as shown in Figure 4.

Obtain initial population: in this model, genetic algorithm is used for seeking optimum evolution rule and is nested in the cellular machine model, and chromosome comprises station and select and two parts of operation ordering, requires height for speed of convergence.Therefore do not adopt the random initializtion method, but directly with min (w _q+ wl ÷ pe) the station selection scheme that obtains for the station choice criteria; (explain: what cell scheduling was studied is the single step scheduling that time and space disperses to reach the workpiece sequencing scheme that obtains according to FCFS rule combination workpiece processing priority; So the workpiece of each research unit can be considered simultaneously and arrives); Station selective staining body in the corresponding respectively coding and operation ordering chromosome, the initial solution of acquisition genetic algorithm.As in the instance 1 according to p ₁-p ₂-p ₃-p ₄-p ₅-p ₆-p ₇-p ₈Order carry out station and select, it is 5-4-5-3-4-3-2-2 that one of them station that obtains is selected initial solution, corresponding device s ₅-s ₄-s ₅-s ₃-s ₄-s ₃-s ₂-s ₂Operation ordering initial solution is 1-2-4-7-5-6-3-8, corresponding workpiece p ₁-p ₂-p ₄-p ₇-p ₅-p ₆-p ₃-p ₈, also be operation O ₁₃-O ₂₂-O ₄₁-O ₇₂-O ₅₃-O ₆₅-O ₃₂-O ₈₇Concrete implementation is illustrated in fig. 5 shown below.

In the flow process, a station time array is set, the equipment time array in the instance 1 just, the accumulation process time that is used to write down each station.Each element value of this array of init state is 0, so unit one is directly selected the shortest equipment process time, with being added to process time of the selected station of this workpiece on the element corresponding in the array.

The chromosomal initial solution of operation confirms according to the time change point in the equipment time array, gets one and is placed on the front when running into two workpiece processing start times when identical, appointing, and this is to not influence of integral body processing ordering.Because the same time of same station can only be processed a workpiece.If two workpiece are identical initial process time, these two workpiece must be positioned at two different processing stations so, even ordering indicates successively, actual is that two workpiece begin processing simultaneously on different stations.

According to said method, the sequencing when the random arrangement workpiece is selected station, 8 workpiece just have 8! , be 40320 kinds of row's methods also, can obtain a plurality of different feasible solutions, form the initial population of genetic algorithm.If the number N=10 of colony randomly draws 10 kinds of workpiece and selects the station order, obtain 10 chromosomes, form the initial solution of instance 1.Simulation process is seen Fig. 6, and the result sees table 6 instance 1 initial solution.

Table 6

Select fitness function: fitness has embodied the objective function of Optimization Model, need meet the following conditions during design:

(1) monodrome, continuous, non-negative, maximization.This condition is easier to realize.

(2) rationally consistance.Require the good and bad degree of fitness value reflection homographic solution, reaching of this condition is often difficult to weigh.

(3) calculated amount is little.The fitness function design should be simple as far as possible, can reduce the complicacy on computing time and the space like this, and reduction assesses the cost.

(4) highly versatile.Fitness should be general as far as possible to certain type of particular problem, preferably need not the user and change the parameter in the fitness function.

According to these 4 conditions, directly objective function is made as fitness function.Have

$\mathrm{fit}\left(F\right)=F=0.4\&times;\frac{\underset{i=1}{\overset{n}{\mathrm{\&Sigma;}}}\min T_{\mathrm{iy}}}{\underset{i=1}{\overset{n}{\mathrm{\&Sigma;}}}\left(T_i{A}_i\right)}+0.3\&times;\frac{\underset{i=1}{\overset{n}{\mathrm{\&Sigma;}}}\left(T_i{A}_i\right)}{\underset{y=1}{\overset{m}{\mathrm{\&Sigma;}}}\left(T(1-{\mathrm{\&lambda;}}_y)\right)}+0.3\&times;\frac{\underset{i=1}{\overset{n}{\mathrm{\&Sigma;}}}\left(T_i{A}_i\right)}{C_{\max }\&times;m}---\left(10\right)$

This paper adopts pro rata fitness to distribute.Pro rata fitness distributes the Monte Carlo method that can be described as selection again, is the possibility of proportion of utilization in its filial generation reservation of probability decision of each ideal adaptation degree.If certain individuals i, its fitness is f _i, then its probability tables that is selected is shown,

$P_i=\frac{f_i}{\underset{i=1}{\overset{M}{\mathrm{\&Sigma;}}}{f}_i}---\left(11\right)$

The roulette back-and-forth method is adopted in system of selection.The ideal adaptation degree is converted in proportion chooses probability, have several individuals just wheel disc to be divided into several sectors.As 10 individuals are arranged in the instance 1, then be divided into 10 sectors.Select because carry out 10 times,, be equivalent to rotate wheel disc No. 10 times so produce 10 random numbers between [0,1], the position of pointer when obtaining No. 10 rotating disks and stopping, pointer rests on a certain sector, and the individuality that this sector is represented is promptly selected.

Use fitness apportion design in proportion, obtain the selection probability of above 10 individuals, see table 7.

Table 7

Corresponding wheel disc is as shown in Figure 6.Because initial solution and nonrandom generation, the fitness of each individuals is all higher, so the area discrepancy of each sector is so obvious not as random initial solution in the wheel disc.

Produce 10 random numbers through function rand (), this random number series and the cumulative probability of calculating acquisition are compared, obtain selected individuality, as shown in table 8.Clearly, the individual selected probability that fitness is high is greater than the low individuality of fitness.Competing in the iteration for the first time, individual 6 and 9 are eliminated, and the substitute is the higher individuality of fitness 1 and 8, and this process is referred to as regeneration.The genetic manipulation that will carry out after the regeneration is for intersecting.

Table 8

Intersect: this paper chromosome is divided into two parts, and first's chromosome station selects part directly to adopt two point of contacts bracketing method, and counterpart intersects in twos.With No. 1 in the instance 1 and No. 3 chromosomes is example, intersects synoptic diagram shown in 7.

The chromosome of second portion---in the workpiece sequencing chromosome because the per pass operation can only occur once, if adopt two chromosomes to intersect in twos, the new chromosome of generation be not identical with parent chromosome be exactly illegal.Therefore, the intersection of operation ordering chromosome dyad is selected the POX bracketing method of Zhang Chaoyong proposition for use.Be provided with parent chromosome Parent1 and Parent2, POX produces child chromosome Children1 and Children2, and idiographic flow is following:

(1) random division workpiece collection 1,2,3 ..., n} is two nonvoid subset J1 and J2;

(2) duplicate workpiece that Parent1 is included in J1 to Children1, Parent2 is included in the workpiece of J1 to Children2, keeps their position;

(3) duplicate workpiece that Parent2 is included in J2 to Children1, Parent1 is included in the workpiece of J2 to Children2, keeps their order.

With No. 1 in the instance 1 and No. 3 chromosomes is example, and the synoptic diagram that intersects is as shown in Figure 8.

Variation: the variation of this paper also is divided into two kinds of situation.The chromosomal variation of first is employed in the gene string of station selective staining body and selects a position at random, concentrates at the station of this operation and selects an integer unequal with it at random, replaces current gene, and separating of obtaining like this can guarantee it is feasible solution.The method that the chromosomal variation of second portion adopts gene location to exchange; Promptly from operation ordering chromosome, select the gene of two positions at random; Its position is exchanged, can guarantee so identical numeral can not occur in same the operation ordering chromosome, also promptly separate feasible.No. 1 chromosome first variation in the instance 1 and second portion variation are respectively shown in Fig. 9 (a) and (b).

The total evolutionary process of genetic algorithm: the simple evolutionary process of a generation is summarised as obtains initial population, fitness function, selection, intersection, mutation operation.Select initial population NP=10, crossover probability P _c=0.6, the variation probability P _m=0.001, the first generation evolutionary process that gets initial population is shown in figure 10.Iterate according to this process then, up to satisfying end condition, the chromosome that obtain this moment can be considered the optimal rules of the cellular automaton evolution that after the genetic algorithm optimizing, obtains.

Claims (2)

1. the dynamic dispatching method based on the flexible job shop of heavy parts of cellular machine is characterized in that: comprise the steps:

1) set up the grid model of cellular machine: whole workshop is set at one comprises improved two-dimensional grid system, each grid is represented a processing stations or buffer memory as a cellular, and a certain moment of each station can only be processed a part; The workpiece of buffer area is waited in line; Set total n the station group of the flexible job shop of said heavy parts, every group of station number m (n) that comprises is only relevant with station group alias n, does not generally wait; The station number that each station group comprises with the buffer memory number of its corresponding cache group mutually;, improved expression arrives the workpiece of system at random

2) cellular state is described design: can represent with following mathematical form according to any buffer memory cellular t+1 of cellular state function

state constantly

$S_{C_{(2n+1)j}}^{t+1}=f[S_{C_{(2n+1)j}}^t,(S_{C_{\left(2n\right)m}}^t,S_{C_{(2n+1)m}}^t,S_{C_{(2n+2)j}}^t)]---\left(1\right)$

In the formula (1):

F---local state transformation rule is the job scheduling rule;

---buffer memory cellular C _{(2n+1) j}All station cellular t of upstream station group state set constantly;

(m ≠ j)---buffer memory cellular C _{(2n+1) j}Place other cellulars of buffer memory group C ( _{2n+1) m}T state set constantly;

---buffer memory cellular C _{(2n+1) j}The station cellular C corresponding in the downstream process group with it _{(2n+2) j}T state constantly,

Station cellular t+1 STA representation constantly is equally, arbitrarily:

$S_{C_{\left(2n\right)j}}^{t+1}=f[S_{C_{\left(2n\right)j}}^t,(S_{C_{(2n-1)j}}^t,S_{C_{2n+1}}^t)]---\left(2\right)$

In the formula (2):

---station cellular C _{(2n) j}In the upstream station group with its corresponding cache cellular C _{(2n-1) j}T state constantly;

---station cellular C _{(2n) j}All buffer memory cellular t of downstream buffer memory group state set constantly,

Buffer memory cellular t status attribute constantly is expressed as:

$S_{\mathrm{cb}}^t(w_{\mathrm{ct}},w_{\mathrm{co}},w_{\mathrm{cl}},\mathrm{lq},w_q)---\left(3\right)$

In the formula (3):

---buffer memory cellular C _{(2n+1) j}At t state constantly;

w _Ct---the buffer memory cellular in the product space total volume, static attribute, the unit of unit;

w _Co---occupied capacity in the buffer memory cellular, dynamic attribute, 0≤w _Co≤w _Ct

w _Cl---the unappropriated capacity of this buffer memory cellular, dynamic attribute, w _Cl=w _Ct-w _Co

Lq---queue length, dynamic attribute, the medium workpiece number to be processed of expression buffer memory cellular;

w _q---the stand-by period, dynamic attribute equals the time that last processing work need be waited in the workpiece formation of buffer memory cellular, the day of unit,

(wl is the corresponding required working ability of operation of part),

Pe---working (machining) efficiency, static attribute is selected to represent operation, gets can accomplish several roads in one day and represent operation as working (machining) efficiency, and the pe majority of each station does not wait;

Station cellular t status attribute constantly is expressed as:

$S_{\mathrm{cs}}^t(\mathrm{st},\mathrm{pe},T,s_s,s_{\mathrm{ct}},s_{\mathrm{co}},s_{\mathrm{cl}})---\left(4\right)$

In the formula (4):

---station cellular C _{(2n) j}At t state constantly;

St---affiliated station group, static attribute, st ∈ { C _2n, n is a positive integer;

The time that T---station can be used to process in dispatching cycle altogether, static attribute, the sky/d of unit;

s _s---station busy-idle condition, dynamic attribute, s _s∈ 0,1, and 2}, 0 free time, 1 is busy, 2 faults;

s _Ct---the total working ability of station in dispatching cycle, static attribute, s _Ct=pe * T, the unit of unit;

s _Co---the occupied working ability of station also is the load of station, dynamic attribute, the workpiece that equals to process He processing, workpiece to be processed in the formation of corresponding buffering cellular, the summation of the working ability value that this three part needs, 0≤s _Co≤s _Ct

s _Cl---the remaining working ability of station cellular, dynamic attribute, s _Cl=s _Ct-s _Co,

3) starting condition and boundary condition are set: workpiece particle t status attribute constantly, and expression formula is following:

$S_p^t(p_t,p_f,\mathrm{np},\mathrm{fst},\mathrm{wl},s_n,{\mathrm{dp},\mathrm{qn},t}_a)---\left(5\right)$

In the formula:

---Status ofParticle, particle p _kAt t state constantly, k ∈ N ⁺, the workpiece numbering;

p _t---Procedures in Total, the operation sum of the required processing of particle, static attribute;

p _f---the process number that Procedures Finished, particle have accomplished, dynamic attribute;

Np---Next Procedure, the next process numbering of particle, dynamic attribute;

Fst---Following Station Type, particle accomplish after the current operation next station group that will get into, dynamic attribute;

Wl---Workload, the working ability of the needed station of this particle next process, dynamic attribute, unit are unit;

s _n---Space Needed, the required occupation space of this particle, static attribute, the w of corresponding buffer area _Cl

Dp---Delivery Priority, workpiece processing priority, static attribute, by the decision at delivery date of workpiece, delivery date, tighter priority was high more, and corresponding dp numerical value is big more, and dp gets positive integer;

Qn---Sequence Number of the Queue, the sequence number in the queue order, dynamic attribute, qn gets positive integer;

t _a---Arrive Time, arrive time of cellular, be divided into the time (also promptly begin process time) of the time that arrives the buffer memory cellular (also promptly begin line up time) and arrival station cellular, dynamic attribute,

The initial sum inlet boundary conditional definition of The model system becomes each particle to get into the beat of system, and export boundary condition is the ability to accept of supply chain downstream factory to product;

4) confirm model self-organization evolution rule: the optimization aim of the flexible job shop dynamic scheduling problem of heavy parts is summarised as 3 points: (1) each procedure completion date early; (2) each station rate of load condensate is high, and is idle few; (3) all station balancing the loads in the same station group reduce three according to the characteristics of flexible job shop dynamic dispatching with the self-organization evolution rule of this model, shown in the table 1,

Table 1

4) model evolution process specifically comprises:

Station is selected: station selective rule R _{P → c}: a certain workpiece arrives, and at first judges busy, the spare time or the malfunction of station in its next target station group; According to the quantity that is in the station of same state, preferentially selecting the working (machining) efficiency height also is the big station of pe numerical value, or can accomplish the station of this operation the earliest, promptly compares (w then _q+ wl ÷ pe) value, the station of the correspondence that selective value is minimum; Judge again whether workpiece can be held by its target station cellular and target cache cellular on time, space, also promptly compare s _ClWith the size of wl, and w _ClAnd s _nSize, if s _Cl>=wl and w _Cl>=s _n, workpiece gets into station, otherwise workpiece temporarily gets into the reserve buffer area, treats that the emulation clock advances the step to rejudge when dispatching one the time again;

Workpiece sequencing: workpiece sequencing rule R _{C → p}The processing priority of main combination FCFS (First Come First Served) principle and workpiece is confirmed the processing sequence of workpiece, at first, and according to due in t _aOrdering, t _aMore little workpiece comes front more; The workpiece size of priority dp relatively in twos forward from the team end then; If the dp value of back workpiece is greater than the dp value of front workpiece; Then queue position and the formation sequence number qn with two workpiece exchanges, and is not less than till the dp value of back workpiece the output queue order up to the dp of front workpiece value; Upgrade the state attribute value of buffer memory cellular and each particle

Task triggers: task triggering rule R _TaActivation need satisfy two conditions: (1) target station cellular is idle, s _s=0; (2) workpiece to be processed is positioned at formation foremost, qn=1, and when condition satisfies, workpiece target approach station, the target station is occupied, and puts it and has much to do s _s=1, buffer memory cellular queue length lq=lq-1, the formation sequence number qn of all particles of waiting in the formation subtracts 1, advances one, produces the processing start time, upgrades t simultaneously _a, arrange the process finishing incident after the completion of processing, idle station, workpiece leaves, and returns executive program at last,

Evolution rule is: station selective rule R _{P → c}Be that to machine with the j procedure be starting point, the j+1 procedure machines the (w during this period of time into terminal point _q+ wl ÷ pe) is criterion, selects (w _q+ wl ÷ pe) minimum station is as the processing stations of j+1 procedure, workpiece sequencing rule R _{C → p}Then combine the decision of workpiece processing priority according to the First Come First Served rule.

2. the dynamic dispatching method of the flexible job shop of a kind of heavy parts based on cellular machine as claimed in claim 1 is characterized in that: in the said step 4), adopt genetic algorithm that the evolution rule of cellular machine is optimized, process is following:

If station group is a unit with its corresponding cache group, each the time step in, the scheduling mechanism of each unit is identical, is converted into static one by one flexible scheduling problem, optimal objective has 3: (1) each procedure completion date early; (2) each station rate of load condensate is high, and is idle few; (3) all station balancing the loads in the same station group, below, utilize genetic algorithm to station selective rule R _{P → c}With workpiece sequencing rule R _{C → p}Encode;

The model description of each scheduling unit of cellular machine in each time step is: n workpiece particle will be processed in the station group that comprises the individual similar station of m; M the corresponding m of a station difference buffer memory; The working (machining) efficiency of each station is different, and each workpiece only needs to accomplish one procedure in this station group, and to require alternative station be that in the station group certain is several or whole to the per pass operation according to actual; The per pass operation is processed the required time and is differed on different stations, establish:

Workpiece particle collection P={p ₁, p ₂, p ₃..., p _n;

Station collection s={s ₁, s ₂, s ₃..., s _m;

Buffer memory collection B={b ₁, b ₂, b ₃..., b _m;

Operation O _IjEach workpiece has only unique one procedure in the j procedure of representing i workpiece, this model investigation scope, also is that the corresponding j of each workpiece has only one;

Matrix T process time of x station group, wherein T _IjxyRepresent that the j procedure of i workpiece processes the needed time on y station of x station group;

The ME temporal information, every rate of breakdown is λ _y

Sub-goal weight w ₁, w ₂, w ₃,

All process steps is the shortest total process time; Promptly

is in order to cooperate all the other two sub-goals; Make it relatively uniform, it is handled as follows, make sub-goal 1 be converted into maximal value; And be unlikely to too to approach 0

$F_1=\frac{\underset{i=1}{\overset{n}{\mathrm{\&Sigma;}}}\min T_{\mathrm{iy}}}{\underset{i=1}{\overset{n}{\mathrm{\&Sigma;}}}\left(T_i{A}_i\right)}y\&Element;\{\mathrm{1,2},...,m\}---\left(6\right)$

In the formula (6), what divide subrepresentation is that all workpiece all select process time the shortest station to add total process time in man-hour, is the minimum value of ideal state; Denominator is represented actual total process time of each feasible solution, 0＜F ₁≤1, T _iBe the row matrix of a series of m of comprising element, the corresponding T of each element _Ijxy, can be abbreviated as T for each cell scheduling _Iy

A _iThen be the column matrix of a series of m of comprising element, each element a _Iy∈ 0,1}, wherein y is corresponding be exactly station sequence number y ∈ 1,2 ... m},

Each station rate of load condensate is high, is converted into all station overall utilization F ₂High:

$F_2=\frac{\underset{i=1}{\overset{n}{\mathrm{\&Sigma;}}}\left(T_i{A}_i\right)}{\underset{y=1}{\overset{m}{\mathrm{\&Sigma;}}}\left(T(1-{\mathrm{\&lambda;}}_y)\right)}---\left(7\right)$

In the formula (7), T is the real time of each simulation step length representative, sets before emulation begins, and does not change in the simulation process; λ _yBe every rate of breakdown in the corresponding station group,

All station balancing the loads in the same station group, balanced ratio F ₃Maximum, according to

has

$F_3=\frac{\underset{i=1}{\overset{n}{\mathrm{\&Sigma;}}}\left(T_i{A}_i\right)}{C_{\max }\&times;m}---\left(8\right)$

In the formula (8), C _MaxFor loading maximum spent process time of station, m is the station sum of station group for this reason,

Through weight add legal each discrete unit that obtains cellular machine each the time scheduling general objective function in the step:

F=max (w ₁F ₁+ w ₂F ₂+ w ₃F ₃), promptly

$F=\max (w_1\&times;\frac{\underset{i=1}{\overset{n}{\mathrm{\&Sigma;}}}\min T_{\mathrm{iy}}}{\underset{i=1}{\overset{n}{\mathrm{\&Sigma;}}}\left(T_i{A}_i\right)}+w_2\&times;\frac{\underset{i=1}{\overset{n}{\mathrm{\&Sigma;}}}\left(T_i{A}_i\right)}{\underset{y=1}{\overset{m}{\mathrm{\&Sigma;}}}\left(T(1-{\mathrm{\&lambda;}}_y)\right)}+w_3\&times;\frac{\underset{i=1}{\overset{n}{\mathrm{\&Sigma;}}}\left(T_i{A}_i\right)}{C_{\max }\&times;m})---\left(9\right)$

Processing constraint condition is following:

1) resource constraint,

A) same equipment synchronization can only be processed a workpiece;

B) the same one procedure of same workpiece can only be by an apparatus processing;

C) the per pass operation of each workpiece is just Once you begin processed and can not be stopped,

2) process constraint,

A) not constraint successively between the operation of different workpieces;

B) constraint is successively arranged between the operation of same workpiece,

3) weight constraints,

A) weight of each sub-goal is between 0 and 1: 0≤w _i≤1i ∈ 1,2,3};

B) the weight sum of all sub-goals is

Obtain initial population: with min (w _q+ wl ÷ pe) the station selection scheme that obtains for the station choice criteria; And, obtain the initial solution of genetic algorithm according to station selective staining body and operation ordering chromosome in corresponding respectively coding of the regular workpiece sequencing scheme that combines workpiece processing priority to obtain of FCFS;

Objective function is made as fitness function, has

$\mathrm{fit}\left(F\right)=F=0.4\&times;\frac{\underset{i=1}{\overset{n}{\mathrm{\&Sigma;}}}\min T_{\mathrm{iy}}}{\underset{i=1}{\overset{n}{\mathrm{\&Sigma;}}}\left(T_i{A}_i\right)}+0.3\&times;\frac{\underset{i=1}{\overset{n}{\mathrm{\&Sigma;}}}\left(T_i{A}_i\right)}{\underset{y=1}{\overset{m}{\mathrm{\&Sigma;}}}\left(T(1-{\mathrm{\&lambda;}}_y)\right)}+0.3\&times;\frac{\underset{i=1}{\overset{n}{\mathrm{\&Sigma;}}}\left(T_i{A}_i\right)}{C_{\max }\&times;m}---\left(10\right)$

Adopt pro rata fitness distribution method, proportion of utilization is in the possibility of its filial generation reservation of probability decision of each ideal adaptation degree, if certain individuals i, its fitness is f _i, then its probability tables that is selected is shown,

$P_i=\frac{f_i}{\underset{i=1}{\overset{M}{\mathrm{\&Sigma;}}}{f}_i}---\left(11\right)$

The roulette back-and-forth method is adopted in system of selection, the ideal adaptation degree is converted in proportion chooses probability, has several individuals just wheel disc to be divided into several sectors;

Intersect: said chromosome is divided into two parts, and first's chromosome station selects part directly to adopt two point of contacts bracketing method, and counterpart intersects in twos; The chromosome of second portion---in the workpiece sequencing chromosome because the per pass operation can only occur once, if adopt two chromosomes to intersect in twos, the new chromosome of generation be not identical with parent chromosome be exactly illegal;

Variation: the chromosomal variation of first, be employed in the gene string of station selective staining body and select a position at random, concentrate at the station of this operation and select an integer unequal at random with it; Replace current gene; Separating of obtaining like this can guarantee it is feasible solution, and the method for gene location exchange is adopted in the chromosomal variation of second portion, promptly from operation ordering chromosome, selects the gene of two positions at random; Its position is exchanged; Can guarantee so identical numeral can not occur in same the operation ordering chromosome, also promptly separate feasible

The simple evolutionary process of a generation comprises obtains initial population, fitness function, selection, intersection, mutation operation; Simple evolutionary process according to a generation iterates then; Up to satisfying end condition, the chromosome that obtain this moment can be considered the optimal rules of the cellular automaton evolution that after the genetic algorithm optimizing, obtains.3, the dynamic dispatching method of the flexible job shop of a kind of heavy parts based on cellular machine as claimed in claim 2; It is characterized in that: in the said interlace operation; The intersection of operation ordering chromosome dyad is selected the POX bracketing method for use; Be provided with parent chromosome Parent1 and Parent2, POX produces child chromosome Children1 and Children2, and idiographic flow is following:

1) random division workpiece collection 1,2,3 ..., n} is two nonvoid subset J1 and J2;

2) duplicate workpiece that Parent1 is included in J1 to Children1, Parent2 is included in the workpiece of J1 to Children2, keeps their position;

3) duplicate workpiece that Parent2 is included in J2 to Children1, Parent1 is included in the workpiece of J2 to Children2, keeps their order.

Images